For a while, we've been grappling with one big question: How do we give agents secure access to services without them reading any secrets? Today, we provide an early answer: Agent Vault, an open source, HTTP credential proxy and vault.
Agents like OpenClaw or Claude Code can proxy requests through Agent Vault regardless of the method an agent uses to interact with any target service: API, CLI, SDK, MCP.
With Agent Vault, we’re rethinking how secrets should be consumed by agents. We believe that vaults and/or secret stores are here to stay but the way in which secrets are delivered to fit the ergonomics of how agents operate will change drastically.
In the current state, agents cannot be trusted with holding secrets directly and so there has to be a dedicated credential broker beside each agent, be it through a dedicated service, sidecar, or egress layer; to securely attach credentials onto every request to the outside world. With this proxy in place, you can inspect proxied requests and, in the future, apply firewall rules to apply restrictions to traffic flowing through the proxy.
The Agent Vault project by
@infisical is an early peek into a trend that we believe many folks including Anthropic, Brex, Browser Use and others have caught onto which is the separation of the agent from its credentials.
Check it out.